5

I've been attempting to find a formula for $\int\tan(x)^{1/n}dx, n\in\Bbb N$.

I started out by performing the substitution $$u=\tan(x)^{1/n}$$
$$dx=\frac {nu^{n-1}}{1+u^{2n}}du$$ to transform the integral into $$n\int \frac{u^n}{1+u^{2n}}du$$ This is a relatively simple rational function, so it should be possible to solve via a partial fraction expansion. For the simple case where $n=2$, I was able to rewrite the integral as $$\frac{\sqrt2}{2}\left(\int\frac{u}{u^2-\sqrt2u+1}du - \int\frac{u}{u^2+\sqrt2u+1}du\right)$$
Which can then be solved by completing the square to yield $$\frac{\sqrt2}{4}\ln\left(\left(\sqrt{\tan{x}}-\frac{\sqrt2}{2}\right)^2+\frac{1}{2}\over\left(\sqrt{\tan{x}}+\frac{\sqrt2}{2}\right)^2+\frac{1}{2}\right)+\frac{\sqrt2}{2}\left(\arctan\left({\sqrt{2\tan{x}}-1}\right)+\arctan\left({\sqrt{2\tan{x}}+1}\right)\right)+C$$
However, I can't figure out how to factor the denominator for the partial fraction expansion of the general case. It will always be a product of $n$ irreducible quadratic factors, clearly, but I don't know how to locate them. I have managed to find the complex linear factors, using Euler's identity to find that they're of the form $$x\pm e^{im\pi\over{2n}}$$ where m is allowed to vary from 1 to $2n$. How can I combine these linear factors into real, irreducible quadratics for use in my partial fraction expansion? Alternatively, is there a simpler method of evaluating this integral? Any help would be vastly appreciated.

Quanto
  • 97,352

3 Answers3

2

With $a_k=\frac{(2k-1)\pi}{2n}$

$$x^{2n}+1= \prod_{k=1}^{2n}(x- e^{i a_k})$$ and the corresponding partial fractional decomposition is

\begin{align} \frac{x^n}{x^{2n}+1} & = -\frac1{2n}\sum_{k=1}^{2n}\frac{e^{i(n+1)a_k}}{x-e^{i a_k}}\\ &=-\frac1{2n}\sum_{k=1}^{n}\left( \frac{e^{i (n+1)a_k}}{x-e^{ia_k} }+ \frac{e^{-i (n+1)a_k}}{x-e^{-ia_k}} \right)\\ &= -\frac1{n}\sum_{k=1}^{n} \frac{x\cos(n+1)a_k+\cos na_k }{x^2-2x\cos a_k +1} \\ &= \frac1{n}\sum_{k=1}^{n} \frac{(-1)^{k+1} x\sin a_k}{x^2-2x\cos a_k +1} \end{align} Then, integrate term-wise to obtain \begin{align} & \int\tan^{\frac1n}(t)\> dt= \int \frac{nx^n}{x^{2n}+1}dx\\ =& \sum_{k=0}^{n}\frac{(-1)^{k+1}}2\left( \sin a_k \ln\left(x^2 - 2x\cos a_k + 1\right) + 2\cos a_k \tan^{-1}\frac{x- \cos a_k}{\sin a_k} \right) \end{align}

Quanto
  • 97,352
1

I do not think that this is an answer but it is too long for a comment.

Adressing this problem, you enter in the world of the gaussian hypergeometric function since $$I_n=n\int \frac{u^n}{1+u^{2n}}\,du=\frac{n \,u^{n+1} }{n+1}\,\, _2F_1\left(1,\frac{n+1}{2 n};\frac{3n+1}{2 n};-u^{2 n}\right)$$ which, as you showed, can simplify for $n=2$.

In other words, back to $x$, $$I_n=\int{\tan^{\frac 1n} (x)}\,dx=\frac{n \tan ^{\frac{n+1}{n}}(x)}{n+1}\,\,\, _2F_1\left(1,\frac{n+1}{2 n};\frac{3n+1}{2 n};-\tan ^2(x)\right)$$

Edit

In your comment, you reported that $$u^6+1=(u^2+\sqrt{3}u+1)(u^2-\sqrt{3}u+1)(u^2+1)$$ So, using partial fraction decomposition $$\frac{u^3}{1+u^6}=\frac{u}{3 \left(u^2+1\right)}+\frac{u}{6 \left(u^2-\sqrt{3} u+1\right)}+\frac{u}{6 \left(u^2+\sqrt{3} u+1\right)}$$ which can be integrated leading to $$\int\frac{u^3}{1+u^6}\,du=-\frac{1}{6} \log \left(u^2+1\right)+\frac{1}{12} \log \left(u^2-\sqrt{3} u+1\right)+\frac{1}{12} \log \left(u \left(u+\sqrt{3}\right)+1\right)-\frac{\tan ^{-1}\left(\sqrt{3}-2 u\right)}{2 \sqrt{3}}-\frac{\tan ^{-1}\left(2 u+\sqrt{3}\right)}{2 \sqrt{3}}$$ Here again, the logarithms can be combined together as well as the arctangents.

Much more tedious is the factorization of $u^8+1$; it leads to $$u^8+1=\left(u^2-2 u \sin \left(\frac{\pi }{8}\right)+1\right) \left(u^2+2 u \sin \left(\frac{\pi }{8}\right)+1\right)$$ $$ \left(u^2-2 u \cos \left(\frac{\pi }{8}\right)+1\right) \left(u^2+2 u \cos \left(\frac{\pi }{8}\right)+1\right)$$ leading to a small monster I shall not reproduce here (too long !).

Similarly, we could write $$u^{10}+1=(1+u^2)(u^8-u^6+u^4-u^2+1)$$ and the last term again can be factorized as the product of four quadratic polynomials.

$$\left(u^2-\sqrt{\frac{1}{2} \left(5-\sqrt{5}\right)} u+1\right) \left(u^2+\sqrt{\frac{1}{2} \left(5-\sqrt{5}\right)} u+1\right)$$ $$ \left(u^2-\sqrt{\frac{1}{2} \left(5+\sqrt{5}\right)} u+1\right) \left(u^2+\sqrt{\frac{1}{2} \left(5+\sqrt{5}\right)} u+1\right)$$

Looking for something else, I found this post where Dr.MV's answer would be very interesting for yur work.

  • I'm afraid that the definition of the gaussian hypergeometric function is too complex for me to fully understand, but thank you for your quick response! Is the first equation a standard identity of some sort, and if not, could you explain how you derived it? Also, is there a way to know which values of $n$ will allow the expression to simplify? I haven't gone through the full calculations, but $x^6+1=(x^2+\sqrt{3}x+1)(x^2-\sqrt{3}x+1)(x^2+1)$, which potentially means that the integral would simplify for $n=3$ as well as $n=2$. I will update my question with my work on that case if it helps. – Andrew Stelzer Nov 04 '16 at 21:30
  • @AndrewStelzer. I was afraid too ! Congratulations for having done it fot $n=3$. Inspired by your solution, I shall try and, may be update my answer. – Claude Leibovici Nov 05 '16 at 03:54
  • My answer actually looks slightly different from yours (I suspect due to a sign error on my part), but I'm more concerned with showing that these integrals can theoretically be found in a relatively simple manner than actually calculating all of them. I'm not sure that I understand how you're factoring the polynomials for the partial fraction expansion though. Is the middle term always built out of $\sin{\frac{\pi}{2n}}$ and $\cos{\frac{\pi}{2n}}$, and if so, how did you derive that form? Also, I'm very curious as to how the hypergeometric function provided a consise solution above. – Andrew Stelzer Nov 06 '16 at 01:58
  • Oh, is it correct to say that the hypergeometric function solution arose from the integral definition given in JJacquelin's answer here? All the definitions for the function that I found previously were based on the rising factorial function, which did not seem related at all to me. Please disregard my last question above if this is where it comes from. :) – Andrew Stelzer Nov 06 '16 at 02:11
1

We are going to the fact that the natural logarithm is a periodic function to find the $2n$ distinct roots of $x^{2n}+1$. Observe that $$\forall z,w\in \mathbb{C},z\neq 0,\ln(z^w)\equiv w\ln(z)\pmod{2\pi i}$$ which is equivalent to $$\forall z,w\in \mathbb{C},z\neq 0,\exists k\in \mathbb{Z},\ln(z^w)=w\ln(z)+2\pi ik$$ Then \begin{align} x^{2n}+1&=0\\ \Leftrightarrow x^{2n}&=-1\\ \end{align} Since $x=0$ is not a root, it follows that \begin{align} x^{2n}&=-1\\ \Leftrightarrow \ln(x^{2n})&=\ln(-1)=\pi i\\ \Leftrightarrow \exists k\in \mathbb{Z},2n\ln(x)&=\pi i+2\pi i k\\ \Leftrightarrow \exists k\in \mathbb{Z},\ln(x)&=\frac{\pi i(1+2k)}{2n}\\ \Leftrightarrow \exists k\in \mathbb{Z},x&=\exp(\frac{\pi i(1+2k)}{2n}) \tag{1} \end{align} Since $\forall m\in\mathbb{Z},\exp(z)=\exp(z+2\pi im)$, we only need the first $2n$ integers in (1), that is $k\in\{1,2,\cdots,2n\}$. Hence, the $2n$ roots of $x^{2n}+1$ are given by \begin{align} r_1&=\exp(\frac{\pi i(1+2(1))}{2n})\\ r_2&=\exp(\frac{\pi i(1+2(2))}{2n})\\ &\vdots\\ r_{2n}&=\exp(\frac{\pi i(1+2(2n))}{2n}) \end{align} An element $\gamma$ in a field K is a root of a polynomial with coefficients in K, iff $x-\gamma$ divides the polynomial, then \begin{align} x^{2n}+1&=\prod_{k=1}^{2n}x-r_k=\prod_{k=1}^{2n}x-\exp(\frac{\pi i(1+2k)}{2n})\\ \end{align} Using Euler's formula $e^{i\theta}=\cos(\theta)+i\sin(\theta)$ \begin{align} x^{2n}+1&=\prod_{k=1}^{2n}x-\Big(\cos(\frac{\pi (1+2k)}{2n})+i\sin(\frac{\pi (1+2k)}{2n})\Big) \end{align} You can finish the integration using partial fraction decomposition. A useful theorem will be the following:
Theorem 1.0 [1] Let $P(x)$ & $Q(x)$ be polynomials of real coefficients where $\deg(P)<\deg(Q)$ & $Q(x)=(x-r_1)(x-r_2)\cdots(x-r_n)$, then $$\frac{P(x)}{Q(x)}=\sum_{i=1}^n\frac{P(r_i)}{Q'(r_i)}\frac{1}{x-r_i}$$ Theorem 1.0 can be used if $n<2n$, that is, if $0<n$. Let $0<n$, observe that $$\int \tan^{\frac{1}{n}}(x)dx=\int \frac{nu^n}{u^{2n}+1}du$$ Let $P(u)=u^n$ & $Q(u)=u^{2n}+1=\prod_{k=1}^{2n}x-\big(\cos(\frac{\pi (1+2k)}{2n})+i\sin(\frac{\pi (1+2k)}{2n})\big)$, then $\deg(P)=n<2n=\deg(Q)$ & $Q'(u)=2nu^{2n-1}$, it follows by Theorem 1.0 that \begin{align} \int \frac{nu^n}{u^{2n}+1}du&=n\int \sum_{k=1}^{2n}\frac{P(r_k)}{Q'(r_k)}\frac{1}{u-r_k}du=n\int \sum_{k=1}^{2n}\frac{(r_k)^{n}}{2n(r_k)^{2n-1}}\frac{1}{u-r_k}du\\ &=\frac{1}{2}\int \sum_{k=1}^{2n}\frac{(r_k)^{n}}{(r_k)^{2n-1}}\frac{1}{u-r_k}du =\frac{1}{2}\sum_{k=1}^{2n}\frac{(r_k)^{n}}{(r_k)^{2n-1}}\int\frac{1}{u-r_k}du\\ &=\frac{1}{2}\sum_{k=1}^{2n}\frac{(r_k)^{n}}{(r_k)^{2n-1}}\ln|u-r_k|+c \end{align} where $r_k$ is the kth root of $u^{2n}+1$. Substituting $r_k=\cos(\frac{\pi (1+2k)}{2n})+i\sin(\frac{\pi (1+2k)}{2n})$ and $u=\tan^{\frac{1}{n}}(x)$ we obtain $$\frac{1}{2}\sum_{k=1}^{2n}\frac{\Big(\cos(\frac{\pi (1+2k)}{2n})+i\sin(\frac{\pi (1+2k)}{2n})\Big)^{n}}{\Big(\cos(\frac{\pi (1+2k)}{2n})+i\sin(\frac{\pi (1+2k)}{2n})\Big)^{2n-1}}\ln|\tan^{\frac{1}{n}}(x)-\cos(\frac{\pi (1+2k)}{2n})-i\sin(\frac{\pi (1+2k)}{2n})|+c$$

[1] https://en.wikipedia.org/wiki/Partial_fraction_decomposition#Procedure